Embed Size (px)
Citation preview
Generating functions for weighted Hurwitz numbers∗
Mathieu Guay-Paquet1 and J. Harnad2,3
1 Universite du Quebec a Montreal201 Av du President-Kennedy, Montreal QC, Canada H2X 3Y7
email: [email protected] Centre de recherches mathematiques, Universite de MontrealC. P. 6128, succ. centre ville, Montreal, QC, Canada H3C 3J7
e-mail: [email protected] Department of Mathematics and Statistics, Concordia University
7141 Sherbrooke W., Montreal, QC Canada H4B 1R6
Abstract
Weighted Hurwitz numbers for n-sheeted branched coverings of the Riemann sphereare introduced, together with associated weighted paths in the Cayley graph of Sn gen-erated by transpositions. The generating functions for these, which include all formerlystudied cases, are 2D Toda τ -functions of generalized hypergeometric type. Two newtypes of weightings are defined by coefficients in the Taylor expansion of the exponen-tiated quantum dilogarithm function. These are shown to provide q-deformations ofstrictly monotonic and weakly monotonic path enumeration generating functions. Thestandard double Hurwitz numbers are recovered from both types in the classical limit.By suitable interpretation of the parameter q, the corresponding statistical mechanicsof random branched covers is related to that of Bose gases.
1 Introduction
In [7, 9], a method was developed for constructing generating functions for different variants
of Hurwitz numbers, counting branched coverings of the Riemann sphere satisfying certain
specified conditions. These were all shown to be 2D Toda τ -functions [19, 21, 20] of the special
hypergeometric type [18]. A natural combinatorial implementation of the construction was
shown to lead to equivalent interpretations of these in terms of path-counting in the Cayley
graph of the symmetric group generated by transpositions. All previously known cases of
generating functions for Hurwitz numbers were placed within this approach, and several
new examples were studied, both from the enumerative geometrical and the combinatorial
viewpoint.
In the present work, this approach is extended to include any 2D Toda τ -function of
hypergeometric type, interpreted as a generating function for some suitably defined problem
∗Work supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) andthe Fonds de recherche du Quebec – Nature et technologies (FRQNT).
1
of weighted enumeration of paths in the Cayley graph of Sn. In several examples, it is shown
that this may equivalently be interpreted as weighted enumeration of branched covers of the
Riemann sphere. The only additional ingredient needed is the specification of a suitably
defined weighting for the branch points, depending on their ramification indices.
It is straightforward to recover all previous cases studied in [17, 5, 7, 9, 22, 2, 1, 11],
and add to these a number of new ones that were never previously considered. One special
case leads to the notion of q-deformed Hurwitz numbers or quantum Hurwitz numbers,
of which we provide two possible variants. These may be seen as q-deformations of the
previously studied generating functions for strictly monotonic and weakly monotonic path
counting, which correspond in the Hurwitz number framework respectively to Belyi curves
[22, 11, 2] and to signed counting, at fixed genus. The classical limit of both these weighted
enumerations is easily seen to be the standard double Hurwitz number generating functions
introduced by Okounkov [17]. In the quantum case the number of branch points is a random
variable, as is the profile of any individual branch point. The τ -function may be viewed as
a generating function for the expectation values of the Hurwitz numbers.
2 Generating functions for hypergeometric τ-functions
2.1 Weight generating functions and Jucys-Murphy elements
Let
G(z) =∞∑k=0
Gkzk (2.1)
be the Taylor series of a complex analytic function in a neighbourhood of the origin, with
G(0) = 1. In what follows, this will be referred to as the weight generating function since,
it will determine the weights associated with path enumeration in the Cayley graph of Sngenerated by the transpositions, and also, in the cases considered, those associated to the
ramification structure of branched covers of the Riemann sphere. Developing further the
methods introduced in [7, 9], we will show how to use such functions to construct generating
functions for weighted Hurwitz numbers that are 2D Toda τ -functions [19, 21, 20]. As with
ordinary Hurwitz numbers, these are also interpretable as weighted enumeration of certain
paths in the Cayley graph of the symmetric group Sn generated by transpositions. We refer
the reader to [7, 9], for details of the construction, notation and further examples.
Let (a, b) ∈ Sn denote the transposition interchanging the elements a and b, and
Jb :=b−1∑a=1
(a, b), b = 1, . . . , n (2.2)
the Jucys-Murphy elements [12, 15, 3] of the group algebra C[Sn], which generate a maximal
commutative subalgebra. We associate an element G(z,J ) of the completion of the center
2
of the group algebra Z(C[Sn])) by forming the product
G(z,J ) :=n∏a=1
G(zJa). (2.3)
It follows that under multiplication such elements determine endomorphisms of Z(C[Sn]))
that are diagonal in the basis {Fλ} of orthogonal idempotents
G(z,J )Fλ = rGλ (z)Fλ, (2.4)
where the eigenvalue is given by the content product formula [18, 7, 9],
rG(z)λ :=
∏(i,j)∈λ
G(z(j − i)) (2.5)
taken over the elements of the Young diagram of the partition λ of weight |λ| = n.
Ref. [7] shows how to use such elements to define 2D Toda τ -functions that serve as
generating functions for combinatorial invariants enumerating certain paths in the Cayley
graph of Sn generated by transpositions. The associated 2D Toda τ -function τG(t, s) is of
hypergeometric type [18], and may be expressed as a diagonal sum over products of Schur
functions
τG(t, s) =∑λ
rG(z)λ Sλ(t)Sλ(s), (2.6)
where
t = (t1, t2, . . . ), t = (s1, s2, . . . ) (2.7)
are the 2D Toda flow variables, which may be identified in this notation with the power sum
symmetric functions
ti = ipi, si = ip′i. (2.8)
(See [14] for notation and definitions involving symmetric functions.)
Remark 2.1. No lattice site N ∈ Z is indicated in (2.6), since in the examples considered below,only N = 0 is required. The N dependence can be introduced by replacing the factor G(z(j− i)) inthe content product formula (2.5) by G(z(N + j− i)), and multiplying by an overall λ-independentfactor rG0 (z) in a standard way [18, 8]. This produces a lattice of τ -functions τG(N, t, s) which,for all the cases considered below, may be explicitly expressed in terms of τG(0, t, s) =: τG(t, s) byapplying a suitable transformation of the parameters involved [17, 9], and a multiplicative factordepending on N .
Using the Frobenius character formula, and the series expansions (2.1), we can obtain an
equivalent expansion in terms of products of power sum symmetric functions and a power
series in z
τG(t, s) =∞∑k=0
∑µ,ν,|µ|=|ν|
F kG(µ, ν)Pµ(t)Pν(s)z
k. (2.9)
3
The coefficients F kG(µ, ν) will be interpreted as weighted enumerations of paths in the Cayley
graph starting at an element in the conjugacy class of cycle type µ to the class of type ν. A
geometrical interpretation of F kG(µ, ν) will also be given as weighted Hurwitz numbers, which
give weighted enumerations of branched covers of the Riemann sphere. Various examples of
such generating functions have been studied in [5, 7, 9], including more complicated cases
that involve dependence on families of auxiliary parameters constructed in a multiplicative
way.
2.2 Fermionic representation
Double KP τ -functions of the form (2.6) have the fermionic representation [18, 7, 8, 9]
τG(t, s) = 〈0|γ+(t)CGγ−(t)|0〉 (2.10)
where, in terms of the fermionic creation and annihilation operators {ψi, ψ†i }i∈Z satisfying
the usual anticommutation relations
[ψi, ψ†j ]+ = δij (2.11)
and vacuum state |0〉 vanishing conditions
ψi|0〉 = 0, for i < 0, ψ†i |0〉 = 0, for i ≥ 0, (2.12)
we have
CG = e∑∞j=−∞ TGj :ψjψ
†j : (2.13)
TGj =
j∑k=1
lnG(zk), TG0 = 0, TG−j = −j−1∑k=0
lnG(−zk) for j > 0, (2.14)
γ+(t) = e∑∞i=1 tiJi , γ−(s) = e
∑∞i=1 siJi , Ji =
∑k∈Z
ψkψ†k+i, i ∈ Z. (2.15)
More generally, using the charge N vacuum state
|N〉 = ψN−1 · · ·ψ0|0〉, | −N〉 = ψ†−N · · ·ψ†−1|0〉, N ∈ N+, (2.16)
we may define a 2D Toda lattice of τ -functions by
τG(N, t, s) = 〈N |γ+(t)CGγ−(t)|N〉 (2.17)
=∑λ
rG(z)λ (N)Sλ(t)Sλ(s), (2.18)
where
rG(z)λ (N) :=
∏(i,j)∈λ
G(z(N + j − i)), (2.19)
and these satisfy the infinite set of Hirota bilinear equations for the 2D Toda lattice hierarchy
[21, 19, 20].
4
2.3 Examples
The following lists some examples that were studied previously in [17, 5, 7, 9], as well
as some further examples that will be considered in the present work, together with their
interpretation as generating functions for the weighted enumeration of various configurations
of branched covers of the Riemann sphere and, correspondingly, weighted paths in the Cayley
graph of the symmetric group.
Example 2.1. Double Hurwitz numbers [17].
G(z) = exp(z) := ez, exp(z,J ) = ez∑nb=1 Jb , expi =
1
i!
rexpj = ezj , rexpλ (z) = ez2
∑`(λ)i=1 λi(λi−2i+1), T exp
j =1
2j(j + 1)z. (2.20)
The coefficients F kG(µ, ν) in this case are Okounkov’s double Hurwitz numbers [17], which enumeratebranched coverings of the Riemann sphere with ramification types µ and ν at 0 and ∞, and kadditional simple branch points. Equivalently, they enumerate d-step paths in the Cayley graph ofSn from an element in the conjugacy class of cycle type µ to the class of cycle type ν. The genusis given by the Riemann-Hurwitz formula
2− 2g = `(µ) + `(ν)− k. (2.21)
Example 2.2. Monotone double Hurwitz numbers [7, 9].
G(z) = E(z) := 1 + z, E(z,J ) =n∏a=1
(1 + zJa)
E1 = 1, Gj = Ej = 0 for j > 1,
rEj = 1 + zj, rEλ (z) =∏
((ij)∈λ
(1 + z(j − i)), (2.22)
TEj =
j∑k=1
ln(1 + kz), TE−j = −j−1∑k=1
ln(1− kz), j > 0. (2.23)
The coefficients F kG(µ, ν) are double Hurwitz numbers for Belyi curves, [22, 11, 2], which enumeraten-sheeted branched coverings of the Riemann sphere having three ramification points, with ramifi-cation profile types µ and ν at 0 and∞, and a single additional branch point, with n−k preimages.The genus is again given by the Riemann-Hurwitz formula (2.21).
Equivalently, they enumerate k-step paths in the Cayley graph of Sn from an element in theconjugacy class of cycle type µ to the class of cycle type ν, that are strictly monotonically increasingin their second elements [7].
Example 2.3. Multimonotone double Hurwitz numbers [9].
G(z) = El(z) := (1 + z)l, El(z,J ) =n∏a=1
(1 + zJa)l, Eli =
(li
)
5
rEl
j = (1 + zj)l, rEl
λ (z) =∏
(ij)∈λ
(1 + z(j − i))l,
TEl
j = l
j∑k=1
ln(1 + kz), TE−j = −lj−1∑k=1
ln(1− kz), j > 0. (2.24)
The coefficients F kEl
(µ, ν) are double Hurwitz numbers that enumerate branched coverings of theRiemann sphere with ramification profile types µ and ν at 0 and∞, and l additional branch points[22, 11, 2], such that the sum of the colengths of their ramification profile type is equal to k.
Equivalently, they enumerate k-step paths in the Cayley graph of Sn, formed from consecutivetranspositions, from an element in the conjugacy class of cycle type µ to the class of cycle typeν, that consist of a sequence of l strictly monotonically increasing subsequences in their secondelements [7, 9].
Example 2.4. Weakly monotone double Hurwitz numbers [5, 7].
G(z) = H(z) :=1
1− z, H(z,J ) =
n∏a=1
(1− zJa)−1, Hi = 1, i ∈ N+
rHj = (1− zj)−1, rHλ (z) =∏
(ij)∈λ
(1− z(j − i))−1, (2.25)
THj = −j∑
k=1
ln(1− kz), TE−j =
j−1∑k=1
ln(1 + kz), j > 0. (2.26)
The coefficients F kH(µ, ν) are double Hurwitz numbers n-sheeted branched coverings of the Riemannsphere curves with branch points at 0 and ∞ having ramification profile types µ and ν, andan arbitrary number of further branch points, such that the sum of the complements of theirramification profile lengths (i.e., the “defect” in the Riemann Hurwitz formula) is equal to k. Thelatter are counted with a sign, which is (−1)n+d times the parity of the number of branch points[9]. The genus is again given by (2.21).
Equivalently, they enumerate k-step paths in the Cayley graph of Sn from an element in theconjugacy class of cycle type µ to the class cycle type ν, that are weakly monotonically increasingin their second elements [7].
6
Example 2.5. Quantum Hurwitz numbers (i).
G(z) =E(q, z) :=∞∏k=0
(1 + qkz) =∞∑k=0
Ek(q)zk, (2.27)
Ei(q) :=
i∏j=0
qj
1− qj, (2.28)
E(q,J ) =
n∏a=1
∞∏i=0
(1 + qizJa), (2.29)
rE(q)j =
∞∏k=0
(1 + qkzj), rE(q)λ (z) =
∞∏k=0
n∏(ij)∈λ
(1 + qkz(j − i)), (2.30)
TE(q)j = −
j∑i=1
Li2(q,−zi), TE(q)−j =
j∑i=0
Li2(q, zi), j > 0. (2.31)
This generating function for weights is related to the quantum dilogarithm function by
E(q, z) = e−Li2(q,−z), Li2(q, z) :=∞∑k=1
zk
k(1− qk). (2.32)
The combinatorial meaning of Ek(q) is: these are generating functions for the number of partitionshaving at most k parts. A slight modification of this is obtained by removing the q0 factor, whichwill be seen below to correspond to removing the zero energy level, giving the generating function
E′(q, z) :=∞∏k=1
(1 + qkz). (2.33)
This will be seen to correspond to a Bose gas with energy levels proportional to the integers.Geometrically, the weighted Hurwitz numbers these generate may be viewed as the expectation
values for weighted covers, in which the weight coincides with that for the Bose gas state whoseenergy levels are identified as proportional to the total ramification defect over each branch point.In the classical limit q → 1, we recover example 2.1
7
Example 2.6. Quantum Hurwitz numbers (ii).
G(z) = H(q, z) :=
∞∏i=0
(1− qiz)−1 = eLi2(q,z) =
∞∑k=0
Hk(q)zk, (2.34)
Hk(q) :=
k∏j=1
q
1− qj, (2.35)
H(q,J ) =∞∏k=0
n∏a=1
(1− qkzJa)−1, (2.36)
rH(q)j =
∞∏k=0
(1− qkzj)−1, rH(q)λ (z) =
∞∏k=0
n∏(ij)∈λ
(1− qkz(j − i))−1. (2.37)
TH(q)j =
j∑i=1
Li2(q, zi), TH(q)−j = −
j−1∑i=1
Li2(q,−zi), j > 0. (2.38)
The modification corresponding to removing the zero energy level state is based similarly on thegenerating function
H ′(q, z) :=
∞∏k=1
(1− qkz)−1. (2.39)
The combinatorial meaning of Hk(q) is: these are generating functions for the number of partitionshaving exactly k parts.
Geometrically, these generate the signed version of the weighted Hurwitz numbers associatedto the Bose gas interpretation, with the sign determined by the parity of the number of branchpoints. In the classical limit q → 1, we again recover example 2.1.
Example 2.7. Double Quantum Hurwitz numbers.
G(z) = Q(q, p, z) := E(q, z)H(p, z) =∞∏i=0
(1 + qiz)(1− piz)−1 =∞∑k=0
Qk(q, p)zk, (2.40)
Qk(q, p) :=
k∑m=0
m∏j=1
qj
1− qj
k−m∏j=1
p
1− pj
, (2.41)
Q(q, p,J ) = E(q, z,J )H(p, z,J ), (2.42)
rQ(q,p)j =
∞∏k=0
1 + qkzj
1− pkzj, r
Q(q,p)λ (z) =
∞∏k=0
n∏(ij)∈λ
1 + qkz(j − i)1− pkz(j − i)
. (2.43)
TQ(q,p)j =
j∑i=1
Li2(p, zi)−j∑i=1
Li2(q,−zi),
TQ(q,p)−j = −
j−1∑i=1
Li2(p,−zi) +
j−1∑i=1
Li2(q, zi), j > 0. (2.44)
Geometrically, these generate a hybrid of two types of counting; i.e., two species of branch points,one of which (“uncoloured”) are counted with the weight corresponding to a Bose gas, the other
8
(“coloured”) are counted with signs, as in the previous example, determined by the parity of thenumber of such branch points. In the classical limit q → 1, p → 1, we again recover example 2.1.The generating partition function Q(q, p, z) for the general case may be identified as the measureused in defining the orthogonality of Macdonald polynomials.
2.4 Weighted expansions
Lemma 2.1. For any weight generating function G(z), we have the following expansion for
G(z,J )
G(z,J ) =∑λ
Gλ Mλ(J )z|λ|, (2.45)
where
Gλ :=
|λ|∏i=1
(Gi)ji =
`(λ)∏j=1
Gλj , (2.46)
with ji the number of parts of λ equal to i, and
Mλ(J ) =∑
b1,...,b`(λ)⊂{1,...,n}
Jb1 . . .Jb`(λ) (2.47)
the monomial sum symmetric function evaluated on the Jucys-Murphy elements.
Proof.
G(z,J ) =n∏a=1
(∞∑k=0
GkzkJ k
a
)
=
(∞∑k1=0
Gk1zk1J k1
1
)· · ·
(∞∑
kn=0
GknzbkJ kn
n
)
=∞∑k=0
zk∑
λ, |λ|=k
∑(b1,...,bk)⊂(1,...,n)
`(λ)∏i=1
GλiJλibi
=∑λ
Gλ Mλ(J )z|λ|. (2.48)
Let {Cµ} denote the basis of the center Z(C[Sn]) of the group algebra C[Sn] consisting
of the cycle sums
Cµ =∑
g∈cyc(µ)
g (2.49)
where cyc(µ) is the conjugacy class of elements whose cycle lengths are given by the partition
µ.
9
Lemma 2.2. Multiplication by Mλ(J ) defines an endomorphism of Z(C[Sn]) which, ex-
pressed in the {Cµ} basis is given by
Mλ(J )Cµ =∑
ν, |ν|=|µ|
|cyc(ν)|−1mλµνCν , (2.50)
where
|cyc(ν)| = |ν|!Zν
, Zν =
|ν|∏i=1
iji(ji)! (2.51)
with ji the number of parts of ν equal to i and mλµν the number of |λ|-step paths (a1b1) · · · (a|λ|b|λ|)
in the Cayley graph of Sn generated by transpositions, starting at an element in the conjugacy
class with cycle type µ and ending in the class ν, that are weakly monotonically increasing,
with λi elements in the successive bands in which the second elements bi remains constant.
Remark 2.2. The enumerative constants mλµν may be interpreted in another way, that is perhaps
more natural, since it puts no restrictions on the monotonicity of the path. If we identify thepartition λ as a signature for a path in the sense that, without any ordering restriction, we counthow many times each particular second element appears in the sequence, and these, in a weaklydescending sequence, are equated to the parts of the partition λ, it is clear that for every sequencethat is weakly increasing in the above sense, with constancy of the second elements in successivesubsequences of lengths equal to the parts of λ, the number of unordered sequences in which suchthe second elements are repeated these numbers of times is
mλµν =
|λ|!∏`(λi=1 λi!
mλµν . (2.52)
It follows that
Proposition 2.3.
G(z,J )Cµ =∞∑k=1
ZνFkG(µ, ν)Cνz
k, (2.53)
where
F kG(µ, ν) =
1
n!
∑λ, |λ|=k
Gλmλµν =
∑λ, |λ|=k
(
`(λ∏i=1
λi!)Gλmλµν (2.54)
is the weighted sum over all such k-step paths, with weight Gλ.
In particular, if G(z) is chosen to be exp := ez, E = 1 + z and H = (1− z)−1, we have
expλ =1∏`(λ)
i=1 λi!, Eλ = δλ,(1)|λ| Hλ = 1, (2.55)
10
and the corresponding weighted combinatorial Hurwitz numbers are
F kexp(µ, ν) =
∑λ, |λ|=k
1∏`(λ)i=1 λi!
mλµν (2.56)
F kE(µ, ν) = m(1)k
µν (2.57)
F kH(µ, ν) =
∑λ, |λ|=k
mλµν . (2.58)
This means that F kexp(µ, ν) is the number of (unordered) sequences of k transpositions leading
from an element in the class of type µ to the class of type ν; F kE(µ, ν) is the number of strictly
monotonically increasing such paths, and F kH(µ, ν) is the number of weakly monotonically
increasing such paths.
If G(z) is chosen to be E(q, z), H(q, z) or Q(q, p, z), respectively, we have
Eλ(q) =
`(λ)∏i=1
Eλi(q), Hλ(q) =
`(λ)∏i=1
Hλi(q) Qλ(q, p) =
`(λ)∏i=1
Qλi(q, p) (2.59)
and the corresponding weighted combinatorial Hurwitz numbers are
F kE(q)(µ, ν) =
1
n!
∑λ, |λ|=k
Eλ(q)mλµν (2.60)
F kH(q)(µ, ν) =
1
n!
∑λ, |λ|=k
Hλ(q)mλµν (2.61)
F kQ(q,p)(µ, ν) =
1
n!
∑λ, |λ|=k
Qλ(q, p)mλµν (2.62)
Their interpretation in terms of weighted branched covers will be given below.
3 The 2D Toda τ-functions τG(z)(t, s)
3.1 Generating functions for weighted paths
For each choice of G(z), we define a corresponding 2D Toda τ -function of generalized hyper-
geometric type (for N = 0) by the formal series
τG(z)(t, s) :=∑λ
rGλ (z)Sλ(t)Sλ(s). (3.1)
It follows from general considerations [20, 8] that this is indeed a double KP τ -function,
that, when extended suitably to a lattice τG(z)(N, t, s) of such τ functions, satisfies the
corresponding system of Hirota bilinear equations of the 2D Toda hierarchy [19, 21, 20].
11
Substituting the Frobenius character formula
Sλ =∑
µ, |µ|=|λ|
Z−1µ χλ(µ)Pµ (3.2)
for each of the factors Sλ(t)Sλ(s) into (3.1) and substituting the corresponding relation
between the bases {Cµ} and {Fλ}
Fλ = h−1λ∑
µ, |µ|=|λ
χλ(µ)Cµ (3.3)
where hλ is the product of the hook lengths of the partitions λ
h−1λ = det
(1
(λi − i+ j)!
), (3.4)
into eqs. (2.4) and (2.53), equating coefficients in the Cµ basis, and using the orthogonality
relation for the characters ∑µ, |µ|=|λ|
χλ(µ)χρ(µ) = Zµδλρ, (3.5)
we obtain the expansion
τG(z)(t, s) =∞∑k=0
∑µ,ν,|µ|=|µ|
zkF kG(µ, ν)Pµ(t)Pν(s). (3.6)
This proves the following result.
Theorem 3.1. τG(z)(t, s) is the generating function for the numbers F kG(µ, ν) of weighted
k-step paths in the Cayley graph, starting at an element in the conjugacy class of cycle type
µ and ending at the conjugacy class of type ν, with weights of all weakly monotonic paths of
type λ given by Gλ.
3.2 The special cases corresponding to examples 2.1 - 2.4
Example 3.2.1. (Sec. 2, example 2.1) In this case G(z) = exp(z),
expi =1
i!, expλ = expλ =
|λ|∏i=1
(i!)ji
−1 =
`(λ)∏j=1
(λi)!
−1 , (3.7)
where ji is the number of parts of λ equal to i,
exp(z,J ) =∞∑k=0
zk
k!
(n∑b=1
Jb
)k(3.8)
=∞∑k=0
zk∑
λ, |λ|=k
`(λ)∏j=1
(λi)!
−1 ∑{b1,...bk}⊂{1,...,n}
J λ1b1 · · · Jλbkbk
(3.9)
12
and
F kexp(µ, ν) =1
n!
∑λ, |λ|=k
`(λ)∏j=1
(λ!)
−1mλµν (3.10)
is the number of k-term products (a1b1) . . . (akbk) such that, if g ∈ cyc(µ), (a1b1) . . . (akbk)g ∈cyc(ν).
Example 3.2.2. (Sec. 2, example 2.2) In this case G(z) = E(z) = 1 + z,
E1 = 1, Ej = 0 for j > 1, Eλ = δλ,(1)|λ| . (3.11)
Therefore ∑λ,|λ|=k
EλMλ(J ) =∑
b1<···<bk
Jb1 · · · Jbk . (3.12)
andF kE(µ, ν) = m(1)k
µν (3.13)
is the number of k-term products (a1b1) . . . (akbk) that are strictly monotonically increasing, suchthat, if g ∈ cyc(µ), (a1b1) . . . (akbk)g ∈ cyc(ν).
Example 3.2.3. (Sec. 2, example 2.3) In this case G(z) = El(z) = (1 + z)l, so
Eli =
(li
), Elλ =
`(λ)∏i=1
(lλi
), λi ≤ l. (3.14)
Therefore
∑λ,|λ|=k
ElλMλ(J ) =∑
λ,|λ|=k
∑(b1,...,b`(λ))⊂(1,...,n)
`(λ)∏i=1
(lλi
)J λibi
= [zk]n∏a=1
(1 + zJa)l (3.15)
where [zk] means the coefficient of zk in the polynomial. and
F kEl(µ, ν) =∑
λ,|λ|=k
`(λ)∏i=1
(lλi
)mλµν (3.16)
is the number of k-term products (a1b1) . . . (akbk) that are such that, if g ∈ cyc(µ), (a1b1) . . . (akbk)g ∈cyc(ν), consisting of l consecutive subsequences, each of which is strictly monotonically increasingin the second elements of each (aibi).
Example 3.2.4. (Sec. 2, example 2.4) In this case G(z) = H(z) = (1− z)−1, so
Gi = 1, ∀i ∈ N+, Gλ = 1 ∀λ. (3.17)
Therefore ∑λ,|λ|=k
GλMλ(J ) =∑
b1≤···≤bk
Jb1 · · · Jbk . (3.18)
13
andF kH(µ, ν) =
∑λ, |λ|=k
mλµν (3.19)
is the number of number of k-term products (a1b1) . . . (akbk) that are weakly monotonically in-creasing, such that, if g ∈ cyc(µ), (a1b1) . . . (akbk)g ∈ cyc(ν).
3.3 Generating functions for Hurwitz numbers: classical countingof covers
The following four examples were previously studied in ref. ([17, 7, 9]. The interpretation of
the associated τ -functions as generating function with Hurwitz numbers of various groups
of branched covers will just be recalled, without proof.
Example 3.3.1. For example 2.1, we have G(z) = exp(z) := ez and the generating τ -function is[17]
τ exp(z)(t, s) =∑λ
ez2
∑`(λ)i=1 λi(λi−2i+1)Sλ(t)Sµ(s)
=∞∑k=0
∑µ,ν, |µ|=|ν|
F kexp(µ, ν)zkPµ(t)Pν(s), (3.20)
where
F kexp(µ, ν) =1
k!H(µ(1) = (2, (1)(n−2)), . . . , µ(k) = (2, (1)(n−2)), µ, ν) (3.21)
is equal to 1k! times Okounkov’s double Hurwitz number; that is, the number of n = |µ| = |ν|
sheeted branched covers with branch points of ramification type µ and ν at the points 0 and ∞,and k further simple branch points.
Example 3.3.2. For example 2.2, we have G(z) = E(z) := 1 + z and the generating τ -functionis [7, 9]
τE(z)(t, s) =∑λ
z|λ|(z−1)λSλ(t)Sµ(s)
=∞∑k=0
zk∑
µ,ν, µ|=|ν|
F kexp(µ, ν)Pµ(t)Pν(s), (3.22)
whereF kE(µ, ν) =
∑µ(1), `∗(µ1)=k
H(µ(1), µ, ν) (3.23)
is now interpreted as the number of n = |µ| = |ν| = |µ(1)| sheeted branched covers with branchpoints of ramification type µ and ν at 0 and ∞, and one further branch point, with colength`∗(µ(1) = 1; i.e. the case of Belyi curves [22, 11, 9, 2] or dessins d’enfants.
14
Example 3.3.3. For example 2.3, we have G(z) = El(z) := (1+z)l and the generating τ -functionis [9] is
τEl(z)(t, s) =
∑λ
z|λ|(z−1)λSλ(t)Sµ(s)
=∞∑k=0
zk∑
µ,ν, |µ|=|ν|=n
F kexp(µ, ν)Pµ(t)Pν(s), (3.24)
whereF kEl(µ, ν) =
∑µ(1),...µ(l)∑li=1
`∗(µi)=k
H(µ(1), . . . µ(l), µ, ν) (3.25)
is now interpreted [9] as the number of n = |µ| = ν| = |µ(i)| sheeted branched covers with branchpoints of ramification type µ and ν at 0 and ∞, and l further branch points, the sums of whosecolengths is k.
Example 3.3.4. For example 2.4, we have G(z) = H(z) = (1−z)−1 and the generating τ -functionis [7, 9]
τH(z)(t, s) =∑λ
(−z)|λ|(−z−1
)λSλ(t)Sµ(s)
=∞∑k=0
zk∑
µ,ν, µ|=|ν|
F kH(µ, ν)Pµ(t)Pν(s) (3.26)
where
F kH(µ, ν) = (−1)n+k∞∑j=1
(−1)j∑
µ(1)...µ(j)∑ji=1
`∗(µi)=k
H(µ(1), . . . µ(j), µ, ν) (3.27)
is now interpreted as the signed counting of n = |µ| = |ν| sheeted branched covers with branchpoints of ramification type µ and ν at 0 and ∞, and any number further branch points, the sum ofwhose colengths is k, with sign determined by the parity of the number of branch points [9].
15
3.4 The τ-functions τE(q,z), τH(q,z) and τQ(q,p,z) as generating func-tions for enumeration of q-weighted paths
The particular cases
τE(q,z)(t, s) :=∑λ
rE(q)λ (z)Sλ(t)Sλ(s) (3.28)
=∞∑k=0
∑µ,ν,|µ|=|ν|
zkF kE(q)(µ, ν)Pµ(t)Pν(s), (3.29)
τH(q,z)(t, s) :=∑λ
rH(q)λ (z)Sλ(t)Sλ(s) (3.30)
=∞∑k=0
∑µ,ν,|µ|=|ν|
zkF kH(q)(µ, ν)Pµ(t)Pν(s), (3.31)
τQ(q,p,z)(t, s) :=∑λ
rQ(q,p)λ (z)Sλ(t)Sλ(s) (3.32)
=∞∑k=0
∑µ,ν,|µ|=|ν|
zkF kQ(q,p)(µ, ν)Pµ(t)Pν(s). (3.33)
may be viewed as special q-deformations of the generating functions associated to Examples
2.1, 2.2 , with G(z) = 1 + z, G(z) = (1 − z)−1 respectively, and the hybrid combination
generated by the product 1+w1−z . The latter were considered previously in [7, 9], and given
both combinatorial and geometrical interpretations in terms of weakly or strictly monotonic
paths in the Cayley graph.
Remark 3.1. Note that, for the special values of the flow parameters (t, s) given by trace invari-ants of a pair of commuting M ×M matrices, X and Y ,
ti =1
itr(Xi), si =
1
itr(Y i) (3.34)
with eigenvalues (x1, . . . , xM ), (y1, . . . , yM ), these may be viewed as special cases of the two typesof basic hypergeometric functions of matrix arguments [6, 18].
3.5 The classical limits of examples 2.5, 2.6, 2.7
Setting q = eε for some small parameter, taking the leading term contribution in the limit
ε→ 0, we obtain
limε→0
E(q, εz) = ez (3.35)
and therefore, taking the scaled limit with z → εz, we obtain
limε→0
τE(q,εz)(t, s) = τ exp(z)(t, s) (3.36)
16
Similarly, we have
limε→0
H(q, εz) = ez (3.37)
and hence
limε→0
τH(q,εz)(t, s) = τ exp(z)(t, s) (3.38)
And finally, for the double quantum Hurwitz case, example 2.7, setting
q = eε, p = eε′
(3.39)
and replacing z by z(1ε
+ 1ε′
), we get
limε,ε′→0
Q(q, p,zεε′
ε+ ε′) = ez (3.40)
and hence
limε,ε′→0
τQ(q,p, zεε′
ε+ε′ )(t, s) = τ exp(z)(t, s). (3.41)
Thus, we recover Okounkov’s classical double Hurwitz number generating function τ exp(z)(t, s)
as the classical limit in each case.
4 Quantum Hurwitz numbers
We now proceed to the interpretation of the quantities F kE(q,z)(µ, ν), F k
H(q,z)(µ, ν) and F kQ(q,p,z)(µ, ν)
as weighted enumerations of branched coverings of the Riemann sphere. The key is the
Frobenius-Schur-Burnside formula [13, Appendix A], [4], expressing the Hurwitz number
H(µ(1), . . . , µ(k)), giving the number of inequivalent n-sheeted branched coverings of the
Riemann sphere, with k distinct branch points whose ramification profiles are given by the
k partitions (µ1, . . . , µk), with weight n in terms of the characters of the symmetric group
H(µ(1), . . . , µ(k)) =∑λ
hk−2λ
k∏i=1
χλ(µ(i))
Zµ(i). (4.1)
4.1 The case E(q). Quantum Hurwitz numbers
It follows from the Frobenius character formula, that the Pochhammer symbol (z)λ may be
written
(z)λ = 1 + hλ∑′
µ, |µ|=|λ|
χλ(µ)
Zµz`∗(µ). (4.2)
where∑′
means the sum over partitions, not including the cycle type of the identity element
(1)|λ|, and
`∗(µ) = |µ| − `(µ) (4.3)
17
is the colength of the partition µ. The content product formula (2.31) for this case may
therefore be written as
rH(q)λ (z) =
∞∏k=0
1 +hλ∑′
µ, |µ|=|λ|χλ(µ)
Zµ(zqk)`
∗(µ)
(4.4)
=∞∑d=0
∞∑k1=0
· · ·∞∑ki=0
∑′
µ(1),...µ(d), |µ(i)|=|λ|∑di=1
`∗(µ(i))=k
d∏i=1
hλχ(µ(i))
Zµ(i)(zqki)`
∗(µ(i)) (4.5)
=∞∑k=0
zk∞∑d=0
∑′
µ(1),...µ(d), |µ(i)|=|λ|∑di=1
`∗(µ(i))=k
d∏i=1
hλχ(µ(i))
Zµ(i)
(1
1− q`∗(µ(i))
), (4.6)
where∑′
means the sum over partitions not including the cycle type of the identity element
(1)|λ| and
`∗(µ) = |µ| − `(µ) (4.7)
is the complement of the length of the partitions µ, which we refer to as the colength.
Substituting this into (3.28) and using the Frobenius character formula (3.2) for each of
the factors Sλ(t)Sλ(s) gives
Theorem 4.1.
τE(q,z)(t, s) =∞∑k=0
zk∑µ,ν|µ|=|ν|
HkE(q)(µ, ν)Pµ(t)Pν(s), (4.8)
where
HkE(q)(µ, ν) :=
∞∑d=0
∑′
µ(1),...µ(d)∑di=1
`∗(µ(i))=k
d∏i=1
(1
1− q`∗(µ(i))
)H(µ(1), . . . , µ(d), µ, ν) (4.9)
are the weighted (quantum) Hurwitz numbers that count the number of branched coverings
with genus g given by (2.21) and sum of colengths k, with weight(
1
1−q`∗(µ(i))
)for every branch
point with colength `∗(µ(i)).
From eq. (3.29 ) it follows that
Corollary 4.2. The weighted (quantum) Hurwitz number for the branched coverings of the
Riemann sphere with genus given by (2.21) is equal to the combinatorial Hurwitz number
given by formula (2.60) enumerating weighted paths in the Cayley graph
HkE(q)(µ, ν) = F k
E(q)(µ, ν). (4.10)
18
4.2 The case H(q). Signed quantum Hurwitz numbers
We proceed similarly for this case. The content product formula (2.38) for this case may be
written as
rH(q)λ (z) =
∞∏k=0
1 +hλ∑′
µ, |µ|=|λ|χλ(µ)
Zµ(−zqk)`∗(µ)
−1
(4.11)
=∞∑k=0
(−z)k∞∑d=0
∑′
µ(1),...µ(d), |µ(i)|=|λ|∑di=1
`∗(µ(i))=k
d∏i=1
hλχ(µ(i))
Zµ(i)
(1
q`∗(µ(i)) − 1
). (4.12)
Substituting this into (3.30) and using the Frobenius character formula (3.2) for each of the
factors Sλ(t)Sλ(s) gives
Theorem 4.3.
τH(q,z)(t, s) =∞∑k=0
zk∑µ,ν|µ|=|ν|
HkH(q)(µ, ν)Pµ(t)Pν(s), (4.13)
where
HkH(q)(µ, ν) := (−1)k
∞∑d=0
∑′
µ(1),...µ(d)∑di=1
`∗(µ(i))=k
d∏i=1
(1
q`∗(µ(i)) − 1
)H(µ(1), . . . , µ(d), µ, ν) (4.14)
are the weighted (quantum) Hurwitz numbers that count the number of branched coverings
with genus g given by (2.21) and sum of colengths k, with weight(
1
q`∗(µ(i))−1
)for every branch
point with colength `∗(µ(i)).
From eq. (3.29) it follows that
Corollary 4.4. The weighted (quantum) Hurwitz number for the branched coverings of the
Riemann sphere with genus given by (2.21) is again equal to the combinatorial Hurwitz
number given by formula (2.61) enumerating weighted paths in the Cayley graph
HkH(q)(µ, ν) = F k
H(q)(µ, ν). (4.15)
4.3 The case Q(q, p). Double Quantum Hurwitz numbers
This case can be understood by combining the results for the previous two multiplicatively.
Since
rQ(q,p)λ (z) = r
E(q)λ (z) r
H(p)λ (z) (4.16)
it follows that
19
Theorem 4.5.
τQ(q,p,z)(t, s) =∞∑k=0
zk∑µ,ν|µ|=|ν|
HkQ(q,p)(µ, ν)Pµ(t)Pν(s), (4.17)
where
HkQ(q,p)(µ, ν) :=
∞∑d=0
∞∑e=0
(−1)∑ej=1 `
∗(ν(j))∑′
µ(1),...µ(d),ν(1),...ν(e)∑di=1
`∗(µ(i))+∑ei=1
`∗(ν(i))=k
d∏i=1
(1
1− q`∗(µ(i))
)
×e∏j=1
(1
p`∗(ν(j)) − 1
)H(µ(1), . . . , µ(d), ν(1), . . . , ν(e), µ, ν) (4.18)
are the weighted (quantum) Hurwitz numbers that count the number of branched coverings
with genus g given by (2.21) and sum of colengths k, with weight(
1
1−q`∗(µ(i))
)for every branch
point of type µ(i) with colength `∗(µ(i)) and weight
((−1)`∗(ν(j))
p`∗(ν(j))−1
)for every branch point of type
ν(j) with colength `∗(ν(j)).
It follows from eq. (3.29) that
Corollary 4.6. The weighted (quantum) Hurwitz number for the branched coverings of the
Riemann sphere with genus given by (2.21) is again equal to the combinatorial Hurwitz
number given by formula (2.62) enumerating weighted paths in the Cayley graph
HkQ(q,p)(µ, ν) = F k
Q(q,p)(µ, ν). (4.19)
4.4 Bose gas model
A slight modification of example 2.4 consists of replacing the generating function E(q, z)
defined in eqs. ( 2.27) by E ′(q, z), as defined in (2.33) and (2.39). The effect of this is simply
to replace the weighting factors 11−q`∗(µ) in eq. (4.9) by 1
q−`∗(µ)−1 .
If we identify
q := e−β~ω0 , β = kT, (4.20)
where ω0 is the lowest frequency excitation in a gas of identical bosonic particles and assume
the energy spectrum of the particles consists of integer multiples of ~ω0
εk = k~ω0, (4.21)
the relative probability of occupying the energy level εk is
qk
1− qk=
e−βεk
1− e−βεk, (4.22)
20
which is the energy distribution of a bosonic gas.
If we associate the branch points to the states of the gas and view the Hurwitz num-
bers H(µ(1), . . . µ(l)) as independent, identically distributed random variables, with the state
energies proportional to the sums over the colengths
ε`∗(µ) = ~`∗(µ)βω0, (4.23)
and weight q`∗(µ(i)
1−q`∗(µ(i), the weighted Hurwitz numbers are given, as in eq. (4.9) by,
HkE′(q)(µ, ν) :=
1
ZkE′(q)
∞∑d=0
∑µ(1),...µ(d)∑di=1
`∗(µ(i))=k
d∏i=1
(q`∗(µ(i))
1− q`∗(µ(i))
)H(µ(1), . . . , µ(d), µ, ν), (4.24)
but normalized by the canonical partition function for fixed total energy k~ω0,
ZkE′(q) :=∞∑d=0
∑µ(1),...µ(d)∑di=1
`∗(µ(i))=k
d∏i=1
(q`∗(µ(i))
1− q`∗(µ(i))
). (4.25)
We may therefore interpret these as expectation values of the Hurwitz numbers associated
to the Fermi gas, and view the corresponding τ -function as a generating function for these
expectation values.
4.5 Multiparameter extensions
By combining these cases multiplicatively, a multiparameter family of generating functions
may be obtained, for which the underlying generator is the product
G(q,w, z) :=l∏
a=1
E(q, wa)m∏b=1
H(q, zb). (4.26)
The resulting τ -functions may be interpreted as basic hypergeometric τ -functions of two
matrix arguments [18]. The interpretation of these multiparametric quantum Hurwitz num-
bers, both in terms of weighted enumeration of branched covers, and weighted paths in the
Cayley graph will be the subject of [10].
Acknowledgements. The authors would like to thank Gaetan Borot and Alexander Orlov for helpful
discussions.
21
References
[1] A. Alexandrov, A.Mironov, A. Morozov and S.Natanzon‘, “On KP-integrable Hurwitz
functions”, arXiv:1405.1395
[2] J. Ambjørn and L. Chekhov, “The matrix model for dessins d’enfants”, arXiv:1404.4240.
[3] P. Diaconis and C. Greene, “Applications of Murphy’s elements”, Stanford Technical
Report 335 (1989).
[4] G. Frobenius, “Uber die Charaktere der symmetrischen Gruppe”, Sitzber. Pruess. Akad.
Berlin, 516-534 (1900).
[5] I. P. Goulden, M. Guay-Paquet and J. Novak, “Toda Equations and Piecewise Polyno-
miality for Mixed Double Hurwitz numbers”, arXiv:1307.2137.
[6] K.I. Gross and D.S. Richards, “Special functions of matrix arguments. I: Algebraic
induction, zonal polynomials, and hypergeometric functions”, Transactions Amer Math
Soc., 301, 781-811 (1987).
[7] Mathieu Guay-Paquet and J. Harnad, “2D Toda τ -functions as combinatorial generating
functions”, arXiv:1405.6303
[8] J. Harnad, A. Yu. Orlov, “Convolution symmetries of integrable hierarchies, matrix
models and tau functions”, in: Integrable Systems, Random Matrices and Random Pro-
cesses, MSRI publications (2014, in press). See also arXiv:0901.0323.
[9] J. Harnad and A. Yu. Orlov, “Hypergeometric τ -functions, Hurwitz numbers and enu-
meration of paths” arXiv:1407.7800.
[10] J. Harnad and A. Yu. Orlov, “Quantum Hurwitz numbers” (in preparation)
[11] M. Kazarian and P. Zograf, “‘Virasoro constraints and topological recursion for
Grothendieck’s dessin counting”, arXiv:1406.5976
[12] A. A. Jucys, “Symmetric polynomials and the center of the symmetric group ring”,
Reports on Mathematical Physics 5(1) (1974), 107–112.
[13] S.K. Lando, A.K. Zvonkin, Graphs on Surfaces and their Applications, Encyclopaedia
of Mathematical Sciences, Volume 141, with appendix by D. Zagier, Springer, N.Y.
(2004).
[14] I.G. Macdonald, Symmetric Functions and Hall Polynomials, Clarendon Press, Oxford,
(1995).
22
[15] G. E. Murphy, “A new construction of Young’s seminormal representation of the sym-
metric groups”, Journal of Algebra 69, 287–297 (1981).
[16] S. M. Natanzon and A. Yu. Orlov, “Hurwitz numbers and BKP hierarchy” Hurwitz
numbers and BKP hierarchy, arXiv:1407.8323 .
[17] A. Okounkov, “Toda equations for Hurwitz numbers”, Math. Res. Lett. 7, 447–453
(2000).
[18] A. Yu. Orlov and D. M. Scherbin, “Hypergeometric solutions of soliton equations”,
Theoretical and Mathematical Physics 128, 906–926 (2001).
[19] K. Takasaki, “Initial value problem for the Toda lattice hierarchy”, in: Group Repre-
sentation and Systems of Differential Equations, 139–163, Adv. Stud. in Pure Math. 4,
(1984).
[20] T. Takebe, “Representation theoretical meaning of the initial value problem for the
Toda lattice hierarchy I”, Lett. Math. Phys. 21 77–84, (1991).
[21] K. Ueno and K. Takasaki, “Toda Lattice Hierarchy”, in:Group Representation and Sys-
tems of Differential Equations, 1–95, Adv. Stud. in Pure Math. 4, (1984).
[22] P. Zograf, “Enumeration of Grothendieck’s dessins and KP hierarchy”, arXiv:1312.2538
(2013).
23
